ctdb/server/ctdb_call.c

   1 /*
   2    ctdb_call protocol code
   3
   4    Copyright (C) Andrew Tridgell  2006
   5
   6    This program is free software; you can redistribute it and/or modify
   7    it under the terms of the GNU General Public License as published by
   8    the Free Software Foundation; either version 3 of the License, or
   9    (at your option) any later version.
  10
  11    This program is distributed in the hope that it will be useful,
  12    but WITHOUT ANY WARRANTY; without even the implied warranty of
  13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14    GNU General Public License for more details.
  15
  16    You should have received a copy of the GNU General Public License
  17    along with this program; if not, see <http://www.gnu.org/licenses/>.
  18 */
  19 /*
  20   see http://wiki.samba.org/index.php/Samba_%26_Clustering for
  21   protocol design and packet details
  22 */
  23 #include "includes.h"
  24 #include "lib/tevent/tevent.h"
  25 #include "lib/tdb/include/tdb.h"
  26 #include "lib/util/dlinklist.h"
  27 #include "system/network.h"
  28 #include "system/filesys.h"
  29 #include "../include/ctdb_private.h"
  30
  31 /*
  32   find the ctdb_db from a db index
  33  */
  34  struct ctdb_db_context *find_ctdb_db(struct ctdb_context *ctdb, uint32_t id)
  35 {
  36         struct ctdb_db_context *ctdb_db;
  37
  38         for (ctdb_db=ctdb->db_list; ctdb_db; ctdb_db=ctdb_db->next) {
  39                 if (ctdb_db->db_id == id) {
  40                         break;
  41                 }
  42         }
  43         return ctdb_db;
  44 }
  45
  46 /*
  47   a varient of input packet that can be used in lock requeue
  48 */
  49 static void ctdb_call_input_pkt(void *p, struct ctdb_req_header *hdr)
  50 {
  51         struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
  52         ctdb_input_pkt(ctdb, hdr);
  53 }
  54
  55
  56 /*
  57   send an error reply
  58 */
  59 static void ctdb_send_error(struct ctdb_context *ctdb,
  60                             struct ctdb_req_header *hdr, uint32_t status,
  61                             const char *fmt, ...) PRINTF_ATTRIBUTE(4,5);
  62 static void ctdb_send_error(struct ctdb_context *ctdb,
  63                             struct ctdb_req_header *hdr, uint32_t status,
  64                             const char *fmt, ...)
  65 {
  66         va_list ap;
  67         struct ctdb_reply_error *r;
  68         char *msg;
  69         int msglen, len;
  70
  71         if (ctdb->methods == NULL) {
  72                 DEBUG(DEBUG_INFO,(__location__ " Failed to send error. Transport is DOWN\n"));
  73                 return;
  74         }
  75
  76         va_start(ap, fmt);
  77         msg = talloc_vasprintf(ctdb, fmt, ap);
  78         if (msg == NULL) {
  79                 ctdb_fatal(ctdb, "Unable to allocate error in ctdb_send_error\n");
  80         }
  81         va_end(ap);
  82
  83         msglen = strlen(msg)+1;
  84         len = offsetof(struct ctdb_reply_error, msg);
  85         r = ctdb_transport_allocate(ctdb, msg, CTDB_REPLY_ERROR, len + msglen,
  86                                     struct ctdb_reply_error);
  87         CTDB_NO_MEMORY_FATAL(ctdb, r);
  88
  89         r->hdr.destnode  = hdr->srcnode;
  90         r->hdr.reqid     = hdr->reqid;
  91         r->status        = status;
  92         r->msglen        = msglen;
  93         memcpy(&r->msg[0], msg, msglen);
  94
  95         ctdb_queue_packet(ctdb, &r->hdr);
  96
  97         talloc_free(msg);
  98 }
  99
 100
 101 /**
 102  * send a redirect reply
 103  *
 104  * The logic behind this function is this:
 105  *
 106  * A client wants to grab a record and sends a CTDB_REQ_CALL packet
 107  * to its local ctdb (ctdb_request_call). If the node is not itself
 108  * the record's DMASTER, it first redirects the packet to  the
 109  * record's LMASTER. The LMASTER then redirects the call packet to
 110  * the current DMASTER. But there is a race: The record may have
 111  * been migrated off the DMASTER while the redirected packet is
 112  * on the wire (or in the local queue). So in case the record has
 113  * migrated off the new destinaton of the call packet, instead of
 114  * going back to the LMASTER to get the new DMASTER, we try to
 115  * reduce rountrips by first chasing the record a couple of times
 116  * before giving up the direct chase and finally going back to the
 117  * LMASTER (again). Note that this works because auf this: When
 118  * a record is migrated off a node, then the new DMASTER is stored
 119  * in the record's copy on the former DMASTER.
 120  *
 121  * The maximum number of attempts for direct chase to make before
 122  * going back to the LMASTER is configurable by the tunable
 123  * "MaxRedirectCount".
 124  */
 125 static void ctdb_call_send_redirect(struct ctdb_context *ctdb,
 126                                     TDB_DATA key,
 127                                     struct ctdb_req_call *c,
 128                                     struct ctdb_ltdb_header *header)
 129 {
 130
 131         uint32_t lmaster = ctdb_lmaster(ctdb, &key);
 132         if (ctdb->pnn == lmaster) {
 133                 c->hdr.destnode = header->dmaster;
 134         } else if ((c->hopcount % ctdb->tunable.max_redirect_count) == 0) {
 135                 c->hdr.destnode = lmaster;
 136         } else {
 137                 c->hdr.destnode = header->dmaster;
 138         }
 139         c->hopcount++;
 140         ctdb_queue_packet(ctdb, &c->hdr);
 141 }
 142
 143
 144 /*
 145   send a dmaster reply
 146
 147   caller must have the chainlock before calling this routine. Caller must be
 148   the lmaster
 149 */
 150 static void ctdb_send_dmaster_reply(struct ctdb_db_context *ctdb_db,
 151                                     struct ctdb_ltdb_header *header,
 152                                     TDB_DATA key, TDB_DATA data,
 153                                     uint32_t new_dmaster,
 154                                     uint32_t reqid)
 155 {
 156         struct ctdb_context *ctdb = ctdb_db->ctdb;
 157         struct ctdb_reply_dmaster *r;
 158         int ret, len;
 159         TALLOC_CTX *tmp_ctx;
 160
 161         if (ctdb->pnn != ctdb_lmaster(ctdb, &key)) {
 162                 DEBUG(DEBUG_ALERT,(__location__ " Caller is not lmaster!\n"));
 163                 return;
 164         }
 165
 166         header->dmaster = new_dmaster;
 167         ret = ctdb_ltdb_store(ctdb_db, key, header, data);
 168         if (ret != 0) {
 169                 ctdb_fatal(ctdb, "ctdb_send_dmaster_reply unable to update dmaster");
 170                 return;
 171         }
 172
 173         if (ctdb->methods == NULL) {
 174                 ctdb_fatal(ctdb, "ctdb_send_dmaster_reply cant update dmaster since transport is down");
 175                 return;
 176         }
 177
 178         /* put the packet on a temporary context, allowing us to safely free
 179            it below even if ctdb_reply_dmaster() has freed it already */
 180         tmp_ctx = talloc_new(ctdb);
 181
 182         /* send the CTDB_REPLY_DMASTER */
 183         len = offsetof(struct ctdb_reply_dmaster, data) + key.dsize + data.dsize + sizeof(uint32_t);
 184         r = ctdb_transport_allocate(ctdb, tmp_ctx, CTDB_REPLY_DMASTER, len,
 185                                     struct ctdb_reply_dmaster);
 186         CTDB_NO_MEMORY_FATAL(ctdb, r);
 187
 188         r->hdr.destnode  = new_dmaster;
 189         r->hdr.reqid     = reqid;
 190         r->rsn           = header->rsn;
 191         r->keylen        = key.dsize;
 192         r->datalen       = data.dsize;
 193         r->db_id         = ctdb_db->db_id;
 194         memcpy(&r->data[0], key.dptr, key.dsize);
 195         memcpy(&r->data[key.dsize], data.dptr, data.dsize);
 196         memcpy(&r->data[key.dsize+data.dsize], &header->flags, sizeof(uint32_t));
 197
 198         ctdb_queue_packet(ctdb, &r->hdr);
 199
 200         talloc_free(tmp_ctx);
 201 }
 202
 203 /*
 204   send a dmaster request (give another node the dmaster for a record)
 205
 206   This is always sent to the lmaster, which ensures that the lmaster
 207   always knows who the dmaster is. The lmaster will then send a
 208   CTDB_REPLY_DMASTER to the new dmaster
 209 */
 210 static void ctdb_call_send_dmaster(struct ctdb_db_context *ctdb_db,
 211                                    struct ctdb_req_call *c,
 212                                    struct ctdb_ltdb_header *header,
 213                                    TDB_DATA *key, TDB_DATA *data)
 214 {
 215         struct ctdb_req_dmaster *r;
 216         struct ctdb_context *ctdb = ctdb_db->ctdb;
 217         int len;
 218         uint32_t lmaster = ctdb_lmaster(ctdb, key);
 219
 220         if (ctdb->methods == NULL) {
 221                 ctdb_fatal(ctdb, "Failed ctdb_call_send_dmaster since transport is down");
 222                 return;
 223         }
 224
 225         if (data->dsize != 0) {
 226                 header->flags |= CTDB_REC_FLAG_MIGRATED_WITH_DATA;
 227         }
 228
 229         if (lmaster == ctdb->pnn) {
 230                 ctdb_send_dmaster_reply(ctdb_db, header, *key, *data,
 231                                         c->hdr.srcnode, c->hdr.reqid);
 232                 return;
 233         }
 234
 235         len = offsetof(struct ctdb_req_dmaster, data) + key->dsize + data->dsize
 236                         + sizeof(uint32_t);
 237         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_DMASTER, len,
 238                                     struct ctdb_req_dmaster);
 239         CTDB_NO_MEMORY_FATAL(ctdb, r);
 240         r->hdr.destnode  = lmaster;
 241         r->hdr.reqid     = c->hdr.reqid;
 242         r->db_id         = c->db_id;
 243         r->rsn           = header->rsn;
 244         r->dmaster       = c->hdr.srcnode;
 245         r->keylen        = key->dsize;
 246         r->datalen       = data->dsize;
 247         memcpy(&r->data[0], key->dptr, key->dsize);
 248         memcpy(&r->data[key->dsize], data->dptr, data->dsize);
 249         memcpy(&r->data[key->dsize + data->dsize], &header->flags, sizeof(uint32_t));
 250
 251         header->dmaster = c->hdr.srcnode;
 252         if (ctdb_ltdb_store(ctdb_db, *key, header, *data) != 0) {
 253                 ctdb_fatal(ctdb, "Failed to store record in ctdb_call_send_dmaster");
 254         }
 255
 256         ctdb_queue_packet(ctdb, &r->hdr);
 257
 258         talloc_free(r);
 259 }
 260
 261 /*
 262   called when a CTDB_REPLY_DMASTER packet comes in, or when the lmaster
 263   gets a CTDB_REQUEST_DMASTER for itself. We become the dmaster.
 264
 265   must be called with the chainlock held. This function releases the chainlock
 266 */
 267 static void ctdb_become_dmaster(struct ctdb_db_context *ctdb_db,
 268                                 struct ctdb_req_header *hdr,
 269                                 TDB_DATA key, TDB_DATA data,
 270                                 uint64_t rsn, uint32_t record_flags)
 271 {
 272         struct ctdb_call_state *state;
 273         struct ctdb_context *ctdb = ctdb_db->ctdb;
 274         struct ctdb_ltdb_header header;
 275         int ret;
 276
 277         DEBUG(DEBUG_DEBUG,("pnn %u dmaster response %08x\n", ctdb->pnn, ctdb_hash(&key)));
 278
 279         ZERO_STRUCT(header);
 280         header.rsn = rsn + 1;
 281         header.dmaster = ctdb->pnn;
 282         header.flags = record_flags;
 283
 284         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 285
 286         if (state) {
 287                 if (state->call->flags & CTDB_CALL_FLAG_VACUUM_MIGRATION) {
 288                         /*
 289                          * We temporarily add the VACUUM_MIGRATED flag to
 290                          * the record flags, so that ctdb_ltdb_store can
 291                          * decide whether the record should be stored or
 292                          * deleted.
 293                          */
 294                         header.flags |= CTDB_REC_FLAG_VACUUM_MIGRATED;
 295                 }
 296         }
 297
 298         if (ctdb_ltdb_store(ctdb_db, key, &header, data) != 0) {
 299                 ctdb_fatal(ctdb, "ctdb_reply_dmaster store failed\n");
 300
 301                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 302                 if (ret != 0) {
 303                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 304                 }
 305                 return;
 306         }
 307
 308
 309         if (state == NULL) {
 310                 DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_become_dmaster from node %u\n",
 311                          ctdb->pnn, hdr->reqid, hdr->srcnode));
 312
 313                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 314                 if (ret != 0) {
 315                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 316                 }
 317                 return;
 318         }
 319
 320         if (key.dsize != state->call->key.dsize || memcmp(key.dptr, state->call->key.dptr, key.dsize)) {
 321                 DEBUG(DEBUG_ERR, ("Got bogus DMASTER packet reqid:%u from node %u. Key does not match key held in matching idr.\n", hdr->reqid, hdr->srcnode));
 322
 323                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 324                 if (ret != 0) {
 325                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 326                 }
 327                 return;
 328         }
 329
 330         if (hdr->reqid != state->reqid) {
 331                 /* we found a record  but it was the wrong one */
 332                 DEBUG(DEBUG_ERR, ("Dropped orphan in ctdb_become_dmaster with reqid:%u\n from node %u", hdr->reqid, hdr->srcnode));
 333
 334                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 335                 if (ret != 0) {
 336                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 337                 }
 338                 return;
 339         }
 340
 341         ctdb_call_local(ctdb_db, state->call, &header, state, &data, true);
 342
 343         ret = ctdb_ltdb_unlock(ctdb_db, state->call->key);
 344         if (ret != 0) {
 345                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 346         }
 347
 348         state->state = CTDB_CALL_DONE;
 349         if (state->async.fn) {
 350                 state->async.fn(state);
 351         }
 352 }
 353
 354
 355
 356 /*
 357   called when a CTDB_REQ_DMASTER packet comes in
 358
 359   this comes into the lmaster for a record when the current dmaster
 360   wants to give up the dmaster role and give it to someone else
 361 */
 362 void ctdb_request_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 363 {
 364         struct ctdb_req_dmaster *c = (struct ctdb_req_dmaster *)hdr;
 365         TDB_DATA key, data, data2;
 366         struct ctdb_ltdb_header header;
 367         struct ctdb_db_context *ctdb_db;
 368         uint32_t record_flags = 0;
 369         size_t len;
 370         int ret;
 371
 372         key.dptr = c->data;
 373         key.dsize = c->keylen;
 374         data.dptr = c->data + c->keylen;
 375         data.dsize = c->datalen;
 376         len = offsetof(struct ctdb_req_dmaster, data) + key.dsize + data.dsize
 377                         + sizeof(uint32_t);
 378         if (len <= c->hdr.length) {
 379                 record_flags = *(uint32_t *)&c->data[c->keylen + c->datalen];
 380         }
 381
 382         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 383         if (!ctdb_db) {
 384                 ctdb_send_error(ctdb, hdr, -1,
 385                                 "Unknown database in request. db_id==0x%08x",
 386                                 c->db_id);
 387                 return;
 388         }
 389
 390         /* fetch the current record */
 391         ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, key, &header, hdr, &data2,
 392                                            ctdb_call_input_pkt, ctdb, False);
 393         if (ret == -1) {
 394                 ctdb_fatal(ctdb, "ctdb_req_dmaster failed to fetch record");
 395                 return;
 396         }
 397         if (ret == -2) {
 398                 DEBUG(DEBUG_INFO,(__location__ " deferring ctdb_request_dmaster\n"));
 399                 return;
 400         }
 401
 402         if (ctdb_lmaster(ctdb, &key) != ctdb->pnn) {
 403                 DEBUG(DEBUG_ALERT,("pnn %u dmaster request to non-lmaster lmaster=%u gen=%u curgen=%u\n",
 404                          ctdb->pnn, ctdb_lmaster(ctdb, &key),
 405                          hdr->generation, ctdb->vnn_map->generation));
 406                 ctdb_fatal(ctdb, "ctdb_req_dmaster to non-lmaster");
 407         }
 408
 409         DEBUG(DEBUG_DEBUG,("pnn %u dmaster request on %08x for %u from %u\n",
 410                  ctdb->pnn, ctdb_hash(&key), c->dmaster, c->hdr.srcnode));
 411
 412         /* its a protocol error if the sending node is not the current dmaster */
 413         if (header.dmaster != hdr->srcnode) {
 414                 DEBUG(DEBUG_ALERT,("pnn %u dmaster request for new-dmaster %u from non-master %u real-dmaster=%u key %08x dbid 0x%08x gen=%u curgen=%u c->rsn=%llu header.rsn=%llu reqid=%u keyval=0x%08x\n",
 415                          ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
 416                          ctdb_db->db_id, hdr->generation, ctdb->vnn_map->generation,
 417                          (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid,
 418                          (key.dsize >= 4)?(*(uint32_t *)key.dptr):0));
 419                 if (header.rsn != 0 || header.dmaster != ctdb->pnn) {
 420                         DEBUG(DEBUG_ERR,("ctdb_req_dmaster from non-master. Force a recovery.\n"));
 421
 422                         ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
 423                         ctdb_ltdb_unlock(ctdb_db, key);
 424                         return;
 425                 }
 426         }
 427
 428         if (header.rsn > c->rsn) {
 429                 DEBUG(DEBUG_ALERT,("pnn %u dmaster request with older RSN new-dmaster %u from %u real-dmaster=%u key %08x dbid 0x%08x gen=%u curgen=%u c->rsn=%llu header.rsn=%llu reqid=%u\n",
 430                          ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
 431                          ctdb_db->db_id, hdr->generation, ctdb->vnn_map->generation,
 432                          (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid));
 433         }
 434
 435         /* use the rsn from the sending node */
 436         header.rsn = c->rsn;
 437
 438         /* store the record flags from the sending node */
 439         header.flags = record_flags;
 440
 441         /* check if the new dmaster is the lmaster, in which case we
 442            skip the dmaster reply */
 443         if (c->dmaster == ctdb->pnn) {
 444                 ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn, record_flags);
 445         } else {
 446                 ctdb_send_dmaster_reply(ctdb_db, &header, key, data, c->dmaster, hdr->reqid);
 447
 448                 ret = ctdb_ltdb_unlock(ctdb_db, key);
 449                 if (ret != 0) {
 450                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 451                 }
 452         }
 453 }
 454
 455
 456 /*
 457   called when a CTDB_REQ_CALL packet comes in
 458 */
 459 void ctdb_request_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 460 {
 461         struct ctdb_req_call *c = (struct ctdb_req_call *)hdr;
 462         TDB_DATA data;
 463         struct ctdb_reply_call *r;
 464         int ret, len;
 465         struct ctdb_ltdb_header header;
 466         struct ctdb_call *call;
 467         struct ctdb_db_context *ctdb_db;
 468
 469         if (ctdb->methods == NULL) {
 470                 DEBUG(DEBUG_INFO,(__location__ " Failed ctdb_request_call. Transport is DOWN\n"));
 471                 return;
 472         }
 473
 474
 475         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 476         if (!ctdb_db) {
 477                 ctdb_send_error(ctdb, hdr, -1,
 478                                 "Unknown database in request. db_id==0x%08x",
 479                                 c->db_id);
 480                 return;
 481         }
 482
 483         call = talloc(hdr, struct ctdb_call);
 484         CTDB_NO_MEMORY_FATAL(ctdb, call);
 485
 486         call->call_id  = c->callid;
 487         call->key.dptr = c->data;
 488         call->key.dsize = c->keylen;
 489         call->call_data.dptr = c->data + c->keylen;
 490         call->call_data.dsize = c->calldatalen;
 491         call->reply_data.dptr  = NULL;
 492         call->reply_data.dsize = 0;
 493
 494         /* determine if we are the dmaster for this key. This also
 495            fetches the record data (if any), thus avoiding a 2nd fetch of the data
 496            if the call will be answered locally */
 497
 498         ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, call->key, &header, hdr, &data,
 499                                            ctdb_call_input_pkt, ctdb, False);
 500         if (ret == -1) {
 501                 ctdb_send_error(ctdb, hdr, ret, "ltdb fetch failed in ctdb_request_call");
 502                 return;
 503         }
 504         if (ret == -2) {
 505                 DEBUG(DEBUG_INFO,(__location__ " deferred ctdb_request_call\n"));
 506                 return;
 507         }
 508
 509         /* Dont do READONLY if we dont have a tracking database */
 510         if ((c->flags & CTDB_WANT_READONLY) && !ctdb_db->readonly) {
 511                 c->flags &= ~CTDB_WANT_READONLY;
 512         }
 513
 514         if (header.flags & CTDB_REC_RO_REVOKE_COMPLETE) {
 515                 header.flags &= ~(CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY|CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_REVOKE_COMPLETE);
 516                 if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
 517                         ctdb_fatal(ctdb, "Failed to write header with cleared REVOKE flag");
 518                 }
 519                 /* and clear out the tracking data */
 520                 if (tdb_delete(ctdb_db->rottdb, call->key) != 0) {
 521                         DEBUG(DEBUG_ERR,(__location__ " Failed to clear out trackingdb record\n"));
 522                 }
 523         }
 524
 525         /* if we are revoking, we must defer all other calls until the revoke
 526          * had completed.
 527          */
 528         if (header.flags & CTDB_REC_RO_REVOKING_READONLY) {
 529                 talloc_free(data.dptr);
 530                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 531
 532                 if (ctdb_add_revoke_deferred_call(ctdb, ctdb_db, call->key, hdr, ctdb_call_input_pkt, ctdb) != 0) {
 533                         ctdb_fatal(ctdb, "Failed to add deferred call for revoke child");
 534                 }
 535                 talloc_free(call);
 536                 return;
 537         }
 538
 539         /* if we are not the dmaster and are not hosting any delegations,
 540            then send a redirect to the requesting node */
 541         if ((header.dmaster != ctdb->pnn)
 542             && (!(header.flags & CTDB_REC_RO_HAVE_DELEGATIONS)) ) {
 543                 talloc_free(data.dptr);
 544                 ctdb_call_send_redirect(ctdb, call->key, c, &header);
 545
 546                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 547                 if (ret != 0) {
 548                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 549                 }
 550                 return;
 551         }
 552
 553         if ( (!(c->flags & CTDB_WANT_READONLY))
 554         && (header.flags & (CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY)) ) {
 555                 header.flags   |= CTDB_REC_RO_REVOKING_READONLY;
 556                 if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
 557                         ctdb_fatal(ctdb, "Failed to store record with HAVE_DELEGATIONS set");
 558                 }
 559                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 560
 561                 if (ctdb_start_revoke_ro_record(ctdb, ctdb_db, call->key, &header, data) != 0) {
 562                         ctdb_fatal(ctdb, "Failed to start record revoke");
 563                 }
 564                 talloc_free(data.dptr);
 565
 566                 if (ctdb_add_revoke_deferred_call(ctdb, ctdb_db, call->key, hdr, ctdb_call_input_pkt, ctdb) != 0) {
 567                         ctdb_fatal(ctdb, "Failed to add deferred call for revoke child");
 568                 }
 569                 talloc_free(call);
 570
 571                 return;
 572         }
 573
 574         /* If this is the first request for delegation. bump rsn and set
 575          * the delegations flag
 576          */
 577         if ((c->flags & CTDB_WANT_READONLY)
 578         &&  (c->callid == CTDB_FETCH_WITH_HEADER_FUNC)
 579         &&  (!(header.flags & CTDB_REC_RO_HAVE_DELEGATIONS))) {
 580                 header.rsn     += 3;
 581                 header.flags   |= CTDB_REC_RO_HAVE_DELEGATIONS;
 582                 if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
 583                         ctdb_fatal(ctdb, "Failed to store record with HAVE_DELEGATIONS set");
 584                 }
 585         }
 586         if ((c->flags & CTDB_WANT_READONLY)
 587         &&  (call->call_id == CTDB_FETCH_WITH_HEADER_FUNC)) {
 588                 TDB_DATA tdata;
 589
 590                 tdata = tdb_fetch(ctdb_db->rottdb, call->key);
 591                 if (ctdb_trackingdb_add_pnn(ctdb, &tdata, c->hdr.srcnode) != 0) {
 592                         ctdb_fatal(ctdb, "Failed to add node to trackingdb");
 593                 }
 594                 if (tdb_store(ctdb_db->rottdb, call->key, tdata, TDB_REPLACE) != 0) {
 595                         ctdb_fatal(ctdb, "Failed to store trackingdb data");
 596                 }
 597                 free(tdata.dptr);
 598
 599                 ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 600                 if (ret != 0) {
 601                         DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 602                 }
 603
 604                 len = offsetof(struct ctdb_reply_call, data) + data.dsize + sizeof(struct ctdb_ltdb_header);
 605                 r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REPLY_CALL, len,
 606                                             struct ctdb_reply_call);
 607                 CTDB_NO_MEMORY_FATAL(ctdb, r);
 608                 r->hdr.destnode  = c->hdr.srcnode;
 609                 r->hdr.reqid     = c->hdr.reqid;
 610                 r->status        = 0;
 611                 r->datalen       = data.dsize + sizeof(struct ctdb_ltdb_header);
 612                 header.rsn      -= 2;
 613                 header.flags   |= CTDB_REC_RO_HAVE_READONLY;
 614                 header.flags   &= ~CTDB_REC_RO_HAVE_DELEGATIONS;
 615                 memcpy(&r->data[0], &header, sizeof(struct ctdb_ltdb_header));
 616
 617                 if (data.dsize) {
 618                         memcpy(&r->data[sizeof(struct ctdb_ltdb_header)], data.dptr, data.dsize);
 619                 }
 620
 621                 ctdb_queue_packet(ctdb, &r->hdr);
 622
 623                 talloc_free(r);
 624                 return;
 625         }
 626
 627         CTDB_UPDATE_STAT(ctdb, max_hop_count, c->hopcount);
 628
 629         /* Try if possible to migrate the record off to the caller node.
 630          * From the clients perspective a fetch of the data is just as
 631          * expensive as a migration.
 632          */
 633         if (c->hdr.srcnode != ctdb->pnn) {
 634                 if (ctdb_db->transaction_active) {
 635                         DEBUG(DEBUG_INFO, (__location__ " refusing migration"
 636                               " of key %s while transaction is active\n",
 637                               (char *)call->key.dptr));
 638                 } else {
 639                         DEBUG(DEBUG_DEBUG,("pnn %u starting migration of %08x to %u\n",
 640                                  ctdb->pnn, ctdb_hash(&(call->key)), c->hdr.srcnode));
 641                         ctdb_call_send_dmaster(ctdb_db, c, &header, &(call->key), &data);
 642                         talloc_free(data.dptr);
 643
 644                         ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 645                         if (ret != 0) {
 646                                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 647                         }
 648                         return;
 649                 }
 650         }
 651
 652         ret = ctdb_call_local(ctdb_db, call, &header, hdr, &data, true);
 653         if (ret != 0) {
 654                 DEBUG(DEBUG_ERR,(__location__ " ctdb_call_local failed\n"));
 655                 call->status = -1;
 656         }
 657
 658         ret = ctdb_ltdb_unlock(ctdb_db, call->key);
 659         if (ret != 0) {
 660                 DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
 661         }
 662
 663         len = offsetof(struct ctdb_reply_call, data) + call->reply_data.dsize;
 664         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REPLY_CALL, len,
 665                                     struct ctdb_reply_call);
 666         CTDB_NO_MEMORY_FATAL(ctdb, r);
 667         r->hdr.destnode  = hdr->srcnode;
 668         r->hdr.reqid     = hdr->reqid;
 669         r->status        = call->status;
 670         r->datalen       = call->reply_data.dsize;
 671         if (call->reply_data.dsize) {
 672                 memcpy(&r->data[0], call->reply_data.dptr, call->reply_data.dsize);
 673         }
 674
 675         ctdb_queue_packet(ctdb, &r->hdr);
 676
 677         talloc_free(r);
 678 }
 679
 680 /*
 681   called when a CTDB_REPLY_CALL packet comes in
 682
 683   This packet comes in response to a CTDB_REQ_CALL request packet. It
 684   contains any reply data from the call
 685 */
 686 void ctdb_reply_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 687 {
 688         struct ctdb_reply_call *c = (struct ctdb_reply_call *)hdr;
 689         struct ctdb_call_state *state;
 690
 691         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 692         if (state == NULL) {
 693                 DEBUG(DEBUG_ERR, (__location__ " reqid %u not found\n", hdr->reqid));
 694                 return;
 695         }
 696
 697         if (hdr->reqid != state->reqid) {
 698                 /* we found a record  but it was the wrong one */
 699                 DEBUG(DEBUG_ERR, ("Dropped orphaned call reply with reqid:%u\n",hdr->reqid));
 700                 return;
 701         }
 702
 703
 704         /* read only delegation processing */
 705         /* If we got a FETCH_WITH_HEADER we should check if this is a ro
 706          * delegation since we may need to update the record header
 707          */
 708         if (state->c->callid == CTDB_FETCH_WITH_HEADER_FUNC) {
 709                 struct ctdb_db_context *ctdb_db = state->ctdb_db;
 710                 struct ctdb_ltdb_header *header = (struct ctdb_ltdb_header *)&c->data[0];
 711                 struct ctdb_ltdb_header oldheader;
 712                 TDB_DATA key, data, olddata;
 713                 int ret;
 714
 715                 if (!(header->flags & CTDB_REC_RO_HAVE_READONLY)) {
 716                         goto finished_ro;
 717                         return;
 718                 }
 719
 720                 key.dsize = state->c->keylen;
 721                 key.dptr  = state->c->data;
 722                 ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr,
 723                                      ctdb_call_input_pkt, ctdb, False);
 724                 if (ret == -2) {
 725                         return;
 726                 }
 727                 if (ret != 0) {
 728                         DEBUG(DEBUG_ERR,(__location__ " Failed to get lock in ctdb_reply_call\n"));
 729                         return;
 730                 }
 731
 732                 ret = ctdb_ltdb_fetch(ctdb_db, key, &oldheader, state, &olddata);
 733                 if (ret != 0) {
 734                         DEBUG(DEBUG_ERR, ("Failed to fetch old record in ctdb_reply_call\n"));
 735                         ctdb_ltdb_unlock(ctdb_db, key);
 736                         goto finished_ro;
 737                 }
 738
 739                 if (header->rsn <= oldheader.rsn) {
 740                         ctdb_ltdb_unlock(ctdb_db, key);
 741                         goto finished_ro;
 742                 }
 743
 744                 if (c->datalen < sizeof(struct ctdb_ltdb_header)) {
 745                         DEBUG(DEBUG_ERR,(__location__ " Got FETCH_WITH_HEADER reply with too little data: %d bytes\n", c->datalen));
 746                         ctdb_ltdb_unlock(ctdb_db, key);
 747                         goto finished_ro;
 748                 }
 749
 750                 data.dsize = c->datalen - sizeof(struct ctdb_ltdb_header);
 751                 data.dptr  = &c->data[sizeof(struct ctdb_ltdb_header)];
 752                 ret = ctdb_ltdb_store(ctdb_db, key, header, data);
 753                 if (ret != 0) {
 754                         DEBUG(DEBUG_ERR, ("Failed to store new record in ctdb_reply_call\n"));
 755                         ctdb_ltdb_unlock(ctdb_db, key);
 756                         goto finished_ro;
 757                 }
 758
 759                 ctdb_ltdb_unlock(ctdb_db, key);
 760         }
 761 finished_ro:
 762
 763         state->call->reply_data.dptr = c->data;
 764         state->call->reply_data.dsize = c->datalen;
 765         state->call->status = c->status;
 766
 767         talloc_steal(state, c);
 768
 769         state->state = CTDB_CALL_DONE;
 770         if (state->async.fn) {
 771                 state->async.fn(state);
 772         }
 773 }
 774
 775
 776 /*
 777   called when a CTDB_REPLY_DMASTER packet comes in
 778
 779   This packet comes in from the lmaster response to a CTDB_REQ_CALL
 780   request packet. It means that the current dmaster wants to give us
 781   the dmaster role
 782 */
 783 void ctdb_reply_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 784 {
 785         struct ctdb_reply_dmaster *c = (struct ctdb_reply_dmaster *)hdr;
 786         struct ctdb_db_context *ctdb_db;
 787         TDB_DATA key, data;
 788         uint32_t record_flags = 0;
 789         size_t len;
 790         int ret;
 791
 792         ctdb_db = find_ctdb_db(ctdb, c->db_id);
 793         if (ctdb_db == NULL) {
 794                 DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_reply_dmaster\n", c->db_id));
 795                 return;
 796         }
 797
 798         key.dptr = c->data;
 799         key.dsize = c->keylen;
 800         data.dptr = &c->data[key.dsize];
 801         data.dsize = c->datalen;
 802         len = offsetof(struct ctdb_reply_dmaster, data) + key.dsize + data.dsize
 803                 + sizeof(uint32_t);
 804         if (len <= c->hdr.length) {
 805                 record_flags = *(uint32_t *)&c->data[c->keylen + c->datalen];
 806         }
 807
 808         ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr,
 809                                      ctdb_call_input_pkt, ctdb, False);
 810         if (ret == -2) {
 811                 return;
 812         }
 813         if (ret != 0) {
 814                 DEBUG(DEBUG_ERR,(__location__ " Failed to get lock in ctdb_reply_dmaster\n"));
 815                 return;
 816         }
 817
 818         ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn, record_flags);
 819 }
 820
 821
 822 /*
 823   called when a CTDB_REPLY_ERROR packet comes in
 824 */
 825 void ctdb_reply_error(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 826 {
 827         struct ctdb_reply_error *c = (struct ctdb_reply_error *)hdr;
 828         struct ctdb_call_state *state;
 829
 830         state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
 831         if (state == NULL) {
 832                 DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_reply_error\n",
 833                          ctdb->pnn, hdr->reqid));
 834                 return;
 835         }
 836
 837         if (hdr->reqid != state->reqid) {
 838                 /* we found a record  but it was the wrong one */
 839                 DEBUG(DEBUG_ERR, ("Dropped orphaned error reply with reqid:%u\n",hdr->reqid));
 840                 return;
 841         }
 842
 843         talloc_steal(state, c);
 844
 845         state->state  = CTDB_CALL_ERROR;
 846         state->errmsg = (char *)c->msg;
 847         if (state->async.fn) {
 848                 state->async.fn(state);
 849         }
 850 }
 851
 852
 853 /*
 854   destroy a ctdb_call
 855 */
 856 static int ctdb_call_destructor(struct ctdb_call_state *state)
 857 {
 858         DLIST_REMOVE(state->ctdb_db->ctdb->pending_calls, state);
 859         ctdb_reqid_remove(state->ctdb_db->ctdb, state->reqid);
 860         return 0;
 861 }
 862
 863
 864 /*
 865   called when a ctdb_call needs to be resent after a reconfigure event
 866 */
 867 static void ctdb_call_resend(struct ctdb_call_state *state)
 868 {
 869         struct ctdb_context *ctdb = state->ctdb_db->ctdb;
 870
 871         state->generation = ctdb->vnn_map->generation;
 872
 873         /* use a new reqid, in case the old reply does eventually come in */
 874         ctdb_reqid_remove(ctdb, state->reqid);
 875         state->reqid = ctdb_reqid_new(ctdb, state);
 876         state->c->hdr.reqid = state->reqid;
 877
 878         /* update the generation count for this request, so its valid with the new vnn_map */
 879         state->c->hdr.generation = state->generation;
 880
 881         /* send the packet to ourselves, it will be redirected appropriately */
 882         state->c->hdr.destnode = ctdb->pnn;
 883
 884         ctdb_queue_packet(ctdb, &state->c->hdr);
 885         DEBUG(DEBUG_NOTICE,("resent ctdb_call\n"));
 886 }
 887
 888 /*
 889   resend all pending calls on recovery
 890  */
 891 void ctdb_call_resend_all(struct ctdb_context *ctdb)
 892 {
 893         struct ctdb_call_state *state, *next;
 894         for (state=ctdb->pending_calls;state;state=next) {
 895                 next = state->next;
 896                 ctdb_call_resend(state);
 897         }
 898 }
 899
 900 /*
 901   this allows the caller to setup a async.fn
 902 */
 903 static void call_local_trigger(struct event_context *ev, struct timed_event *te,
 904                        struct timeval t, void *private_data)
 905 {
 906         struct ctdb_call_state *state = talloc_get_type(private_data, struct ctdb_call_state);
 907         if (state->async.fn) {
 908                 state->async.fn(state);
 909         }
 910 }
 911
 912
 913 /*
 914   construct an event driven local ctdb_call
 915
 916   this is used so that locally processed ctdb_call requests are processed
 917   in an event driven manner
 918 */
 919 struct ctdb_call_state *ctdb_call_local_send(struct ctdb_db_context *ctdb_db,
 920                                              struct ctdb_call *call,
 921                                              struct ctdb_ltdb_header *header,
 922                                              TDB_DATA *data)
 923 {
 924         struct ctdb_call_state *state;
 925         struct ctdb_context *ctdb = ctdb_db->ctdb;
 926         int ret;
 927
 928         state = talloc_zero(ctdb_db, struct ctdb_call_state);
 929         CTDB_NO_MEMORY_NULL(ctdb, state);
 930
 931         talloc_steal(state, data->dptr);
 932
 933         state->state = CTDB_CALL_DONE;
 934         state->call  = talloc(state, struct ctdb_call);
 935         CTDB_NO_MEMORY_NULL(ctdb, state->call);
 936         *(state->call) = *call;
 937         state->ctdb_db = ctdb_db;
 938
 939         ret = ctdb_call_local(ctdb_db, state->call, header, state, data, true);
 940         if (ret != 0) {
 941                 DEBUG(DEBUG_DEBUG,("ctdb_call_local() failed, ignoring return code %d\n", ret));
 942         }
 943
 944         event_add_timed(ctdb->ev, state, timeval_zero(), call_local_trigger, state);
 945
 946         return state;
 947 }
 948
 949
 950 /*
 951   make a remote ctdb call - async send. Called in daemon context.
 952
 953   This constructs a ctdb_call request and queues it for processing.
 954   This call never blocks.
 955 */
 956 struct ctdb_call_state *ctdb_daemon_call_send_remote(struct ctdb_db_context *ctdb_db,
 957                                                      struct ctdb_call *call,
 958                                                      struct ctdb_ltdb_header *header)
 959 {
 960         uint32_t len;
 961         struct ctdb_call_state *state;
 962         struct ctdb_context *ctdb = ctdb_db->ctdb;
 963
 964         if (ctdb->methods == NULL) {
 965                 DEBUG(DEBUG_INFO,(__location__ " Failed send packet. Transport is down\n"));
 966                 return NULL;
 967         }
 968
 969         state = talloc_zero(ctdb_db, struct ctdb_call_state);
 970         CTDB_NO_MEMORY_NULL(ctdb, state);
 971         state->call = talloc(state, struct ctdb_call);
 972         CTDB_NO_MEMORY_NULL(ctdb, state->call);
 973
 974         state->reqid = ctdb_reqid_new(ctdb, state);
 975         state->ctdb_db = ctdb_db;
 976         talloc_set_destructor(state, ctdb_call_destructor);
 977
 978         len = offsetof(struct ctdb_req_call, data) + call->key.dsize + call->call_data.dsize;
 979         state->c = ctdb_transport_allocate(ctdb, state, CTDB_REQ_CALL, len,
 980                                            struct ctdb_req_call);
 981         CTDB_NO_MEMORY_NULL(ctdb, state->c);
 982         state->c->hdr.destnode  = header->dmaster;
 983
 984         /* this limits us to 16k outstanding messages - not unreasonable */
 985         state->c->hdr.reqid     = state->reqid;
 986         state->c->flags         = call->flags;
 987         state->c->db_id         = ctdb_db->db_id;
 988         state->c->callid        = call->call_id;
 989         state->c->hopcount      = 0;
 990         state->c->keylen        = call->key.dsize;
 991         state->c->calldatalen   = call->call_data.dsize;
 992         memcpy(&state->c->data[0], call->key.dptr, call->key.dsize);
 993         memcpy(&state->c->data[call->key.dsize],
 994                call->call_data.dptr, call->call_data.dsize);
 995         *(state->call)              = *call;
 996         state->call->call_data.dptr = &state->c->data[call->key.dsize];
 997         state->call->key.dptr       = &state->c->data[0];
 998
 999         state->state  = CTDB_CALL_WAIT;
1000         state->generation = ctdb->vnn_map->generation;
1001
1002         DLIST_ADD(ctdb->pending_calls, state);
1003
1004         ctdb_queue_packet(ctdb, &state->c->hdr);
1005
1006         return state;
1007 }
1008
1009 /*
1010   make a remote ctdb call - async recv - called in daemon context
1011
1012   This is called when the program wants to wait for a ctdb_call to complete and get the
1013   results. This call will block unless the call has already completed.
1014 */
1015 int ctdb_daemon_call_recv(struct ctdb_call_state *state, struct ctdb_call *call)
1016 {
1017         while (state->state < CTDB_CALL_DONE) {
1018                 event_loop_once(state->ctdb_db->ctdb->ev);
1019         }
1020         if (state->state != CTDB_CALL_DONE) {
1021                 ctdb_set_error(state->ctdb_db->ctdb, "%s", state->errmsg);
1022                 talloc_free(state);
1023                 return -1;
1024         }
1025
1026         if (state->call->reply_data.dsize) {
1027                 call->reply_data.dptr = talloc_memdup(call,
1028                                                       state->call->reply_data.dptr,
1029                                                       state->call->reply_data.dsize);
1030                 call->reply_data.dsize = state->call->reply_data.dsize;
1031         } else {
1032                 call->reply_data.dptr = NULL;
1033                 call->reply_data.dsize = 0;
1034         }
1035         call->status = state->call->status;
1036         talloc_free(state);
1037         return 0;
1038 }
1039
1040
1041 /*
1042    send a keepalive packet to the other node
1043 */
1044 void ctdb_send_keepalive(struct ctdb_context *ctdb, uint32_t destnode)
1045 {
1046         struct ctdb_req_keepalive *r;
1047
1048         if (ctdb->methods == NULL) {
1049                 DEBUG(DEBUG_INFO,(__location__ " Failed to send keepalive. Transport is DOWN\n"));
1050                 return;
1051         }
1052
1053         r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_KEEPALIVE,
1054                                     sizeof(struct ctdb_req_keepalive),
1055                                     struct ctdb_req_keepalive);
1056         CTDB_NO_MEMORY_FATAL(ctdb, r);
1057         r->hdr.destnode  = destnode;
1058         r->hdr.reqid     = 0;
1059
1060         CTDB_INCREMENT_STAT(ctdb, keepalive_packets_sent);
1061
1062         ctdb_queue_packet(ctdb, &r->hdr);
1063
1064         talloc_free(r);
1065 }
1066
1067
1068
1069 struct revokechild_deferred_call {
1070         struct ctdb_context *ctdb;
1071         struct ctdb_req_header *hdr;
1072         deferred_requeue_fn fn;
1073         void *ctx;
1074 };
1075
1076 struct revokechild_handle {
1077         struct revokechild_handle *next, *prev;
1078         struct ctdb_context *ctdb;
1079         struct ctdb_db_context *ctdb_db;
1080         struct fd_event *fde;
1081         int status;
1082         int fd[2];
1083         pid_t child;
1084         TDB_DATA key;
1085 };
1086
1087 struct revokechild_requeue_handle {
1088         struct ctdb_context *ctdb;
1089         struct ctdb_req_header *hdr;
1090         deferred_requeue_fn fn;
1091         void *ctx;
1092 };
1093
1094 static void deferred_call_requeue(struct event_context *ev, struct timed_event *te,
1095                        struct timeval t, void *private_data)
1096 {
1097         struct revokechild_requeue_handle *requeue_handle = talloc_get_type(private_data, struct revokechild_requeue_handle);
1098
1099         requeue_handle->fn(requeue_handle->ctx, requeue_handle->hdr);
1100         talloc_free(requeue_handle);
1101 }
1102
1103 static int deferred_call_destructor(struct revokechild_deferred_call *deferred_call)
1104 {
1105         struct ctdb_context *ctdb = deferred_call->ctdb;
1106         struct revokechild_requeue_handle *requeue_handle = talloc(ctdb, struct revokechild_requeue_handle);
1107         struct ctdb_req_call *c = (struct ctdb_req_call *)deferred_call->hdr;
1108
1109         requeue_handle->ctdb = ctdb;
1110         requeue_handle->hdr  = deferred_call->hdr;
1111         requeue_handle->fn   = deferred_call->fn;
1112         requeue_handle->ctx  = deferred_call->ctx;
1113         talloc_steal(requeue_handle, requeue_handle->hdr);
1114
1115         /* when revoking, any READONLY requests have 1 second grace to let read/write finish first */
1116         event_add_timed(ctdb->ev, requeue_handle, timeval_current_ofs(c->flags & CTDB_WANT_READONLY ? 1 : 0, 0), deferred_call_requeue, requeue_handle);
1117
1118         return 0;
1119 }
1120
1121
1122 static int revokechild_destructor(struct revokechild_handle *rc)
1123 {
1124         if (rc->fde != NULL) {
1125                 talloc_free(rc->fde);
1126         }
1127
1128         if (rc->fd[0] != -1) {
1129                 close(rc->fd[0]);
1130         }
1131         if (rc->fd[1] != -1) {
1132                 close(rc->fd[1]);
1133         }
1134         kill(rc->child, SIGKILL);
1135
1136         DLIST_REMOVE(rc->ctdb_db->revokechild_active, rc);
1137         return 0;
1138 }
1139
1140 static void revokechild_handler(struct event_context *ev, struct fd_event *fde,
1141                              uint16_t flags, void *private_data)
1142 {
1143         struct revokechild_handle *rc = talloc_get_type(private_data,
1144                                                      struct revokechild_handle);
1145         int ret;
1146         char c;
1147
1148         ret = read(rc->fd[0], &c, 1);
1149         if (ret != 1) {
1150                 DEBUG(DEBUG_ERR,("Failed to read status from revokechild. errno:%d\n", errno));
1151                 rc->status = -1;
1152                 talloc_free(rc);
1153                 return;
1154         }
1155         if (c != 0) {
1156                 DEBUG(DEBUG_ERR,("revokechild returned failure. status:%d\n", c));
1157                 rc->status = -1;
1158                 talloc_free(rc);
1159                 return;
1160         }
1161
1162         talloc_free(rc);
1163 }
1164
1165 struct ctdb_revoke_state {
1166         struct ctdb_db_context *ctdb_db;
1167         TDB_DATA key;
1168         struct ctdb_ltdb_header *header;
1169         TDB_DATA data;
1170         int count;
1171         int status;
1172         int finished;
1173 };
1174
1175 static void update_record_cb(struct ctdb_client_control_state *state)
1176 {
1177         struct ctdb_revoke_state *revoke_state;
1178         int ret;
1179         int32_t res;
1180
1181         if (state == NULL) {
1182                 return;
1183         }
1184         revoke_state = state->async.private_data;
1185
1186         state->async.fn = NULL;
1187         ret = ctdb_control_recv(state->ctdb, state, state, NULL, &res, NULL);
1188         if ((ret != 0) || (res != 0)) {
1189                 DEBUG(DEBUG_ERR,("Recv for revoke update record failed ret:%d res:%d\n", ret, res));
1190                 revoke_state->status = -1;
1191         }
1192
1193         revoke_state->count--;
1194         if (revoke_state->count <= 0) {
1195                 revoke_state->finished = 1;
1196         }
1197 }
1198
1199 static void revoke_send_cb(struct ctdb_context *ctdb, uint32_t pnn, void *private_data)
1200 {
1201         struct ctdb_revoke_state *revoke_state = private_data;
1202         struct ctdb_client_control_state *state;
1203
1204         state = ctdb_ctrl_updaterecord_send(ctdb, revoke_state, timeval_current_ofs(5,0), pnn, revoke_state->ctdb_db, revoke_state->key, revoke_state->header, revoke_state->data);
1205         if (state == NULL) {
1206                 DEBUG(DEBUG_ERR,("Failure to send update record to revoke readonly delegation\n"));
1207                 revoke_state->status = -1;
1208                 return;
1209         }
1210         state->async.fn           = update_record_cb;
1211         state->async.private_data = revoke_state;
1212
1213         revoke_state->count++;
1214
1215 }
1216
1217 static void ctdb_revoke_timeout_handler(struct event_context *ev, struct timed_event *te,
1218                               struct timeval yt, void *private_data)
1219 {
1220         struct ctdb_revoke_state *state = private_data;
1221
1222         DEBUG(DEBUG_ERR,("Timed out waiting for revoke to finish\n"));
1223         state->finished = 1;
1224         state->status   = -1;
1225 }
1226
1227 static int ctdb_revoke_all_delegations(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, TDB_DATA tdata, TDB_DATA key, struct ctdb_ltdb_header *header, TDB_DATA data)
1228 {
1229         struct ctdb_revoke_state *state = talloc_zero(ctdb, struct ctdb_revoke_state);
1230         int status;
1231
1232         state->ctdb_db = ctdb_db;
1233         state->key     = key;
1234         state->header  = header;
1235         state->data    = data;
1236
1237         ctdb_trackingdb_traverse(ctdb, tdata, revoke_send_cb, state);
1238
1239         event_add_timed(ctdb->ev, state, timeval_current_ofs(5, 0), ctdb_revoke_timeout_handler, state);
1240
1241         while (state->finished == 0) {
1242                 event_loop_once(ctdb->ev);
1243         }
1244
1245         status = state->status;
1246
1247         if (status == 0) {
1248                 struct ctdb_ltdb_header new_header;
1249                 TDB_DATA new_data;
1250
1251                 if (ctdb_ltdb_lock(ctdb_db, key) != 0) {
1252                         DEBUG(DEBUG_ERR,("Failed to chainlock the database in revokechild\n"));
1253                         talloc_free(state);
1254                         return -1;
1255                 }
1256                 if (ctdb_ltdb_fetch(ctdb_db, key, &new_header, state, &new_data) != 0) {
1257                         ctdb_ltdb_unlock(ctdb_db, key);
1258                         DEBUG(DEBUG_ERR,("Failed for fetch tdb record in revokechild\n"));
1259                         talloc_free(state);
1260                         return -1;
1261                 }
1262                 header->rsn++;
1263                 if (new_header.rsn > header->rsn) {
1264                         ctdb_ltdb_unlock(ctdb_db, key);
1265                         DEBUG(DEBUG_ERR,("RSN too high in tdb record in revokechild\n"));
1266                         talloc_free(state);
1267                         return -1;
1268                 }
1269                 if ( (new_header.flags & (CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS)) != (CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS) ) {
1270                         ctdb_ltdb_unlock(ctdb_db, key);
1271                         DEBUG(DEBUG_ERR,("Flags are wrong in tdb record in revokechild\n"));
1272                         talloc_free(state);
1273                         return -1;
1274                 }
1275                 new_header.rsn++;
1276                 new_header.flags |= CTDB_REC_RO_REVOKE_COMPLETE;
1277                 if (ctdb_ltdb_store(ctdb_db, key, &new_header, new_data) != 0) {
1278                         ctdb_ltdb_unlock(ctdb_db, key);
1279                         DEBUG(DEBUG_ERR,("Failed to write new record in revokechild\n"));
1280                         talloc_free(state);
1281                         return -1;
1282                 }
1283                 ctdb_ltdb_unlock(ctdb_db, key);
1284         }
1285
1286         talloc_free(state);
1287         return status;
1288 }
1289
1290
1291 int ctdb_start_revoke_ro_record(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, TDB_DATA key, struct ctdb_ltdb_header *header, TDB_DATA data)
1292 {
1293         TDB_DATA tdata;
1294         struct revokechild_handle *rc;
1295         pid_t parent = getpid();
1296         int ret;
1297
1298         header->flags &= ~(CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY);
1299         header->rsn   -= 1;
1300
1301         if ((rc = talloc_zero(ctdb_db, struct revokechild_handle)) == NULL) {
1302                 DEBUG(DEBUG_ERR,("Failed to allocate revokechild_handle\n"));
1303                 return -1;
1304         }
1305
1306         tdata = tdb_fetch(ctdb_db->rottdb, key);
1307         if (tdata.dsize > 0) {
1308                 uint8_t *tmp;
1309
1310                 tmp = tdata.dptr;
1311                 tdata.dptr = talloc_memdup(rc, tdata.dptr, tdata.dsize);
1312                 free(tmp);
1313         }
1314
1315         rc->status    = 0;
1316         rc->ctdb      = ctdb;
1317         rc->ctdb_db   = ctdb_db;
1318         rc->fd[0]     = -1;
1319         rc->fd[1]     = -1;
1320
1321         talloc_set_destructor(rc, revokechild_destructor);
1322
1323         rc->key.dsize = key.dsize;
1324         rc->key.dptr  = talloc_memdup(rc, key.dptr, key.dsize);
1325         if (rc->key.dptr == NULL) {
1326                 DEBUG(DEBUG_ERR,("Failed to allocate key for revokechild_handle\n"));
1327                 talloc_free(rc);
1328                 return -1;
1329         }
1330
1331         ret = pipe(rc->fd);
1332         if (ret != 0) {
1333                 DEBUG(DEBUG_ERR,("Failed to allocate key for revokechild_handle\n"));
1334                 talloc_free(rc);
1335                 return -1;
1336         }
1337
1338
1339         rc->child = ctdb_fork(ctdb);
1340         if (rc->child == (pid_t)-1) {
1341                 DEBUG(DEBUG_ERR,("Failed to fork child for revokechild\n"));
1342                 talloc_free(rc);
1343                 return -1;
1344         }
1345
1346         if (rc->child == 0) {
1347                 char c = 0;
1348                 close(rc->fd[0]);
1349                 debug_extra = talloc_asprintf(NULL, "revokechild-%s:", ctdb_db->db_name);
1350
1351                 if (switch_from_server_to_client(ctdb, "revokechild-%s", ctdb_db->db_name) != 0) {
1352                         DEBUG(DEBUG_ERR,("Failed to switch from server to client for revokechild process\n"));
1353                         c = 1;
1354                         goto child_finished;
1355                 }
1356
1357                 c = ctdb_revoke_all_delegations(ctdb, ctdb_db, tdata, key, header, data);
1358
1359 child_finished:
1360                 write(rc->fd[1], &c, 1);
1361                 /* make sure we die when our parent dies */
1362                 while (kill(parent, 0) == 0 || errno != ESRCH) {
1363                         sleep(5);
1364                 }
1365                 _exit(0);
1366         }
1367
1368         close(rc->fd[1]);
1369         rc->fd[1] = -1;
1370         set_close_on_exec(rc->fd[0]);
1371
1372         /* This is an active revokechild child process */
1373         DLIST_ADD_END(ctdb_db->revokechild_active, rc, NULL);
1374
1375         rc->fde = event_add_fd(ctdb->ev, rc, rc->fd[0],
1376                                    EVENT_FD_READ, revokechild_handler,
1377                                    (void *)rc);
1378         if (rc->fde == NULL) {
1379                 DEBUG(DEBUG_ERR,("Failed to set up fd event for revokechild process\n"));
1380                 talloc_free(rc);
1381         }
1382         tevent_fd_set_auto_close(rc->fde);
1383
1384         return 0;
1385 }
1386
1387 int ctdb_add_revoke_deferred_call(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, TDB_DATA key, struct ctdb_req_header *hdr, deferred_requeue_fn fn, void *call_context)
1388 {
1389         struct revokechild_handle *rc;
1390         struct revokechild_deferred_call *deferred_call;
1391
1392         for (rc = ctdb_db->revokechild_active; rc; rc = rc->next) {
1393                 if (rc->key.dsize == 0) {
1394                         continue;
1395                 }
1396                 if (rc->key.dsize != key.dsize) {
1397                         continue;
1398                 }
1399                 if (!memcmp(rc->key.dptr, key.dptr, key.dsize)) {
1400                         break;
1401                 }
1402         }
1403
1404         if (rc == NULL) {
1405                 DEBUG(DEBUG_ERR,("Failed to add deferred call to revoke list. revoke structure not found\n"));
1406                 return -1;
1407         }
1408
1409         deferred_call = talloc(rc, struct revokechild_deferred_call);
1410         if (deferred_call == NULL) {
1411                 DEBUG(DEBUG_ERR,("Failed to allocate deferred call structure for revoking record\n"));
1412                 return -1;
1413         }
1414
1415         deferred_call->ctdb = ctdb;
1416         deferred_call->hdr  = hdr;
1417         deferred_call->fn   = fn;
1418         deferred_call->ctx  = call_context;
1419
1420         talloc_set_destructor(deferred_call, deferred_call_destructor);
1421         talloc_steal(deferred_call, hdr);
1422
1423         return 0;
1424 }